The European Biological Variation Study (EuBIVAS): Biological Variation Data for Coagulation Markers Estimated by a Bayesian Model.

BACKGROUND: For biological variation (BV) data to be safely used, data must be reliable and relevant to the population in which they are applied. We used samples from the European Biological Variation Study (EuBIVAS) to determine BV of coagulation markers by a Bayesian model robust to extreme observations and used the derived within-participant BV estimates [CVP(i)] to assess the applicability of the BV estimates in clinical practice. METHOD: Plasma samples were drawn from 92 healthy individuals for 10 consecutive weeks at 6 European laboratories and analyzed in duplicate for activated partial thromboplastin time (APTT), prothrombin time (PT), fibrinogen, D-dimer, antithrombin (AT), protein C, protein S free, and factor VIII (FVIII). A Bayesian model with Student t likelihoods for samples and replicates was applied to derive CVP(i) and predicted BV estimates with 95% credibility intervals. RESULTS: For all markers except D-dimer, CVP(i) were homogeneously distributed in the overall study population or in subgroups. Mean within-subject estimates (CVI) were <5% for APTT, PT, AT, and protein S free, <10% for protein C and FVIII, and <12% for fibrinogen. For APTT, protein C, and protein S free, estimates were significantly lower in men than in women ≤50 years. CONCLUSION: For most coagulation markers, a common CVI estimate for men and women is applicable, whereas for APTT, protein C, and protein S free, sex-specific reference change values should be applied. The use of a Bayesian model to deliver individual CVP(i) allows for improved interpretation and application of the data.

A Bayesian Approach to Biological Variation Analysis.

BACKGROUND: Biological variation (BV) data have many applications for diagnosing and monitoring disease. The standard statistical approaches for estimating BV are sensitive to "noisy data" and assume homogeneity of within-participant CV. Prior knowledge about BV is mostly ignored. The aims of this study were to develop Bayesian models to calculate BV that (a) are robust to "noisy data," (b) allow heterogeneity in the within-participant CVs, and (c) take advantage of prior knowledge. METHOD: We explored Bayesian models with different degrees of robustness using adaptive Student t distributions instead of the normal distributions and when the possibility of heterogeneity of the within-participant CV was allowed. Results were compared to more standard approaches using chloride and triglyceride data from the European Biological Variation Study. RESULTS: Using the most robust Bayesian approach on a raw data set gave results comparable to a standard approach with outlier assessments and removal. The posterior distribution of the fitted model gives access to credible intervals for all parameters that can be used to assess reliability. Reliable and relevant priors proved valuable for prediction. CONCLUSIONS: The recommended Bayesian approach gives a clear picture of the degree of heterogeneity, and the ability to crudely estimate personal within-participant CVs can be used to explore relevant subgroups. Because BV experiments are expensive and time-consuming, prior knowledge and estimates should be considered of high value and applied accordingly. By including reliable prior knowledge, precise estimates are possible even with small data sets.

Biological variation: Evaluation of methods for constructing confidence intervals for estimates of within-person biological variation for different distributions of the within-person effect.

BACKGROUND: Precise estimates of the within-person biological variation, CVI, can be essential both for monitoring patients and for setting analytical performance specifications. The confidence interval, CI, may be used to evaluate the reliability of an estimate, as it is a good measure of the uncertainty of the estimated CVI. The aim of the present study is to evaluate and establish methods for constructing a CI with the correct coverage probability and non-cover probability when estimating CVI. METHOD: Data based on 3 models for distributions for the within-person effect were simulated to assess the performance of 3 methods for constructing confidence intervals; the formula based method for the nested ANOVA, the percentile bootstrap and the bootstrap-t methods. RESULTS: The performance of the evaluated methods for constructing a CI varied, both dependent on the size of the CVI and the type of distributions. The bootstrap-t CI have good and stable performance for the models evaluated, while the formula based are more distribution dependent. The percentile bootstrap performs poorly. CONCLUSION: CI is an essential part of estimation of the within-person biological variation. Good coverage probability and non-cover probabilities for CI are achievable by using the bootstrap-t combined with CV-ANOVA. Supplemental R-code is provided online.

Biological Variation: The Effect of Different Distributions on Estimated Within-Person Variation and Reference Change Values.

BACKGROUND: Good estimates of within-person biological variation, CVI, are essential for diagnosing and monitoring patients and for setting analytical performance specifications. The aim of the present study was to use computer simulations to evaluate the impact of various measurement distributions on different methods for estimating CVI and reference change value (RCV). METHOD: Data were simulated on the basis of 3 models for distributions of the within-person effect. We evaluated 3 different methods for estimating CVI: standard ANOVA, ln-ANOVA, and CV-ANOVA, and 3 different methods for calculating RCV: classic, ln-RCV, and a nonparametric method. We estimated CVI and RCV with the different methods and compared the results with the true values. RESULTS: The performance of the methods varied, depending on both the size of the CVI and the type of distributions. The CV-ANOVA model performed well for the estimation of CVI with all simulated data. The ln-RCV method performed best if data were ln-normal distributed or CVI was less than approximately 12%. The nonparametric RCV method performed well for all simulated data but was less precise. CONCLUSIONS: The CV-ANOVA model is recommended for both calculation of CVI and the step-by-step approach of checking for outliers and homogeneity in replicates and samples. The standard method for calculation of RCV should not be used when using CVs.